Identifying Dominant Drivers of Wind Erosion and Assessing Ecological Risk on the Qinghai-Tibet Plateau Using an XGBoost-SHAP Model

¹ College of Life Science, Qinghai Normal University, Xining 810008, China
² Qinghai Provincial Key Laboratory of Physical Geography and Environmental Process, College of Geographical Science, Qinghai Normal University, Xining 810008, China
³ Key Laboratory of Tibetan Plateau Land Surface Processes and Ecological Conservation (Ministry of Education), Qinghai Normal University, Xining 810008, China
⁴ Academy of Plateau Science and Sustainability, People’s Government of Qinghai Province and Beijing Normal University, Xining 810008, China
⁵ National Positioning Observation and Research Station of Qinghai Lake Wetland Ecosystem, Xining 810008, China

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Wind erosion is a key pathway of land degradation in the alpine drylands of the Qinghai-Tibet Plateau, yet its dominant controls and risk trajectories remain difficult to attribute in a spatially heterogeneous system. Here we integrated the RWEQ with an interpretable machine learning framework (XGBoost coupled with SHAP) to quantify spatiotemporal patterns of potential wind erosion from 1981 to 2020, identify dominant drivers, and delineate management-oriented ecological risk zones. The Plateau-wide mean potential wind erosion modulus was 3.03x104 t/km², with localized hotspots up to 1.08x105 t/km². Wind erosion showed strong seasonality, with most losses occurring from late winter to spring, and a predominantly declining long-term tendency that was spatially heterogeneous across climate classes. SHAP interpretation ranked erosive wind speed as the leading control (mean(|SHAP|) = 10.28), followed by sand content (5.17) and fractional vegetation cover (4.05), with elevation and soil organic matter as secondary modifiers. Dominant driver mapping attributed 59.69% of the Plateau to wind dominance, 9.96% to sand dominance, and 4.49% to vegetation dominance, while 24.74% was non-erosive. Risk zoning indicated extensive recovery areas (54.01%) and substantial gradually worsening belts (17.73%), with rapid worsening hotspots concentrated in wind driven zones (3.33%). The proposed RWEQ plus XGBoost-SHAP framework provides a transparent basis for prioritizing monitoring and targeted mitigation under climate variability.